> De: "Henri Tremblay" <[email protected]>
> À: "mechanical-sympathy" <[email protected]>
> Envoyé: Lundi 22 Mars 2021 12:17:23
> Objet: Re: Tools for understanding GPUs?

> I learned playing with Aparapi ( [ https://aparapi.com/ | 
> https://aparapi.com/ ]
> ).
> It's not super complicated but you need to jump right in mechanical sympathy
> with GPU. To decide vector length and type of memory used. It makes a huge
> difference.

> I would love to tell you to try Java 16 and the Vector API but it doesn't seem
> to support GPU. Just AVX. That's my understanding at least.

yes, 
The Vector API support AVX, AVX-2 and AVX-512 and also NEON but do not support 
GPUs. 

Rémi 

> On Sun, 21 Mar 2021 at 14:36, William Pietri < [ mailto:[email protected] |
> [email protected] ] > wrote:

>> Years of coding, performance tuning, and ops work have given me some level of
>> mechanical sympathy, albeit nothing like many of the contributors to this 
>> list.
>> But now dealing with GPUs I feel like a noob again. 8704 cores! And I feel 
>> like
>> I'm peering through a keyhole trying to figure out what's going on in there.

>> These days I'm working on the Anti-Defamation League's Online Hate Index. We 
>> are
>> creating ML models to spot many kinds of hate, and we aim to apply them at
>> scale to high-volume sources like social media. Luckily, ML experts are
>> handling the modeling. I'm mainly working on the "at scale" part. As a
>> non-profit, optimizing for cost is especially important to us, so I really 
>> need
>> to understand the performance.

>> What tools and resources do folks here like for developing their intuitions
>> about GPU activity? And if you have ML-specific resources, even better.

>> Thanks,

>> William
>> --
>> You received this message because you are subscribed to the Google Groups
>> "mechanical-sympathy" group.
>> To unsubscribe from this group and stop receiving emails from it, send an 
>> email
>> to [ mailto:[email protected] |
>> [email protected] ] .
>> To view this discussion on the web, visit [
>> https://groups.google.com/d/msgid/mechanical-sympathy/895470ac-05fc-c2f5-526f-66b33d5cd65d%40scissor.com?utm_medium=email&utm_source=footer
>> |
>> https://groups.google.com/d/msgid/mechanical-sympathy/895470ac-05fc-c2f5-526f-66b33d5cd65d%40scissor.com
>> ] .

> --
> You received this message because you are subscribed to the Google Groups
> "mechanical-sympathy" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email
> to [ mailto:[email protected] |
> [email protected] ] .
> To view this discussion on the web, visit [
> https://groups.google.com/d/msgid/mechanical-sympathy/CADZL2%3DtNxpPjTCx6v5V5kgmWG09H2g2y_ffk-%2Bmq5N3K8Nmm%2BQ%40mail.gmail.com?utm_medium=email&utm_source=footer
> |
> https://groups.google.com/d/msgid/mechanical-sympathy/CADZL2%3DtNxpPjTCx6v5V5kgmWG09H2g2y_ffk-%2Bmq5N3K8Nmm%2BQ%40mail.gmail.com
> ] .

-- 
You received this message because you are subscribed to the Google Groups 
"mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web, visit 
https://groups.google.com/d/msgid/mechanical-sympathy/756072569.728250.1616419011146.JavaMail.zimbra%40u-pem.fr.

Reply via email to